Cooperative perception is essential for enhancing the perceptual capabilities of intelligent transportation systems. However, existing methods predominantly focus on homogeneous agents, whereas real-world scenarios often involve agents equipped with diverse sensor types. This diversity leads to significant discrepancies in the representation and informational content of shared heterogeneous features, making direct fusion susceptible to information loss or redundancy. To address this challenge, we propose a novel heterogeneous cooperative perception method based on a cross-modality dual-attention mechanism. This mechanism combines global cross-modality attention with local dense spatial attention, capturing inter-modal interactions and fine-grained spatial correlations for robust fusion of heterogeneous features. Additionally, we introduce a multi-scale feature distillation strategy that employs LiDAR bird’s-eye-view features from multiple agents to guide cross-modal alignment, transferring valuable information to image features and reducing modality discrepancies. Quantitative and qualitative experiments conducted on the OPV2V and DAIR-V2X datasets demonstrate that the proposed method achieves state-of-the-art performance in heterogeneous cooperative perception tasks. The framework shows strong potential for deployment in real-world intelligent transportation systems, exhibiting superior detection accuracy, robustness, and computational efficiency, thereby validating its effectiveness and advancement.

Open-pit mining areas have special geological structures, complex road networks, multi-rotation sections and poor road conditions, which bring many challenges to the operation of mining autonomous transport vehicles. Due to the large size and high control difficulty of mining autonomous transport vehicles, poor control effect and inaccurate steering mechanism implementation are prone to occur during the driving process, which leads to the vehicle departure from the reference path. To ensure the safety of autonomous operation in intelligent mine, this paper proposes a method for predicting driving departure of mining autonomous vehicles based on Gated Recurrent Unit (GRU). Firstly, the vehicle's historical trajectory data is obtained via on-board sensors, followed by data cleaning and normalization processes. Key features are extracted, and a vehicle driving scene recognition module based on a GRU-based network is designed using deep learning. Subsequently, a GA-seq2seqGRU trajectory prediction module is constructed utilizing the scene recognition results, and the Genetic Algorithms (GA) is employed to tune the network hyper-parameters. Based on the predicted trajectories, the overall departure risk is calculated using two departure judgment methods based on cross-lane time and predicted lateral deviation, which are mapped to the departure level. Simulation experiments demonstrate that the accuracy of the driving scene recognition model in the proposed method in this paper reaches 0.9721, which is better than the 0.9585 of the Long Short Memory Neural Network (LSTM) model, and the trajectory prediction model based on the results of the driving scene recognition has a smaller RMSE value than the LSTM, GRU, and GA-seq2seqLSTM models when the prediction time domains are 1s, 2s, 3s, 4s, and 5s, and the departure detection module The average time consumed is 0.142ms.

In this study, we tackle the path-tracking problem of a nonlinear four-wheel steering vehicle dynamics model subject to model mismatches and propose a model predictive control (MPC) algorithm based on an event-triggered mechanism (ET -MPC). The goal is to maintain closed-loop control performance while reducing the computational and communication burdens of traditional MPC. We introduce an ET -MPC framework utilizing a model-free reinforcement learning agent with proximal policy optimization (PPO). This agent interacts with the MPC system, progressively learning to determine the optimal event-triggered mechanism. To enhance exploration and training efficiency, we incorporate the Long Short-Term Memory (LSTM) technique into PPO. Experimental results show that the proposed ET -MPC framework, combined with reinforcement learning for reward optimization, demonstrates superior overall performance in path-tracking control of four-wheel steering vehicles.

Autonomous vehicles in mining areas undertake substantial production tasks and are prone to various faults during operation. Early detection of abnormalities, along with timely fault warnings and diagnoses, can enhance transportation safety and increase vehicle turnout rates. This study utilizes driving data from autonomous vehicles in mining areas and considers the characteristics of unstructured road scenes. The driving area is segmented into distinct intervals, and Kullback-Leibler (KL) divergence is applied within each interval to detect anomalies in the vehicle's lateral deviation during operation. Experimental results demonstrate that the proposed method achieves an anomaly detection accuracy of 91.4%, with a false negative rate of 8.3% and a false positive rate of 8.7%.